Comparative Study on Logit and BGEVA Models in Assessing Corporate Default Risk: Based on Large Data of American Listed Companies

Abstract

Using the large financial data of 5926 American listed companies spanning from 1990 to 2015, this paper compares the performance of logit model and binary generalized extreme value additive (BGEVA) model in assessing corporate default risk, during which, only four regression variables including trailing S&P 500 return, US interest rate, firm’s trailing stock return and firm’s distance to default are used. By the comparisons between the estimated cumulative default values with the actual ones per year, we find that BGEVA model performs better in default estimation. Furthermore, along with the decrease of default number, logit model could underestimate the default risk, while BGEVA model maintains a relatively robust estimate, showing an obvious advantage in assessing default of rare events.